Skip to content

What is Syllabus?

Syllabus is the package manager for your agent skills and rules.

Most teams today use skills to guide their agents, but skills aren’t always the ideal package for delivering guidance, and managing skills across repos, organizations, and your local environment can be tricky.

Syllabus solves these problems with two key insights:

  1. Skills need a finer-grained companion: rules.
  2. Skills and rules both need a package manager.

Skills are coarse-grained: a single skill often bundles many insights, which makes it hard to update, version, and understand. A rule is fine-grained: one directive, such as “Never write secrets to logs,” in its own Markdown file. Skills cite the rules they apply instead of restating them.

Syllabus manages both skills and rules the way a package manager manages code. Each project declares the skills and rules it uses and where they come from: public or private libraries, your organization’s shared library, or the project itself. Skills can also come from any Git repository of skill folders, even one that doesn’t use Syllabus.

Syllabus records each skill’s and rule’s source and version, updates them only when you choose, and installs skills for Claude Code, Codex, Cursor, and other agents at once.

With Syllabus, you can:

  • Start every new repo with exactly the right skills and rules in place.
  • Update a skill once in your organization’s library, and have every repo pick up the change on its next update.
  • Document which third-party skills your organization requires, recommends, or makes available for consideration, and have projects automatically install them.
  • Override skills and rules on a per-project basis as needed.

The bigger payoff comes after an agent session. When an agent misses something you expected, run a quick retro: capture the lesson as a rule, and add it to your project or your organization’s library. From then on, every agent gets that guidance when the work calls for it, so the same mistake doesn’t come back. Retros also show where an existing rule or skill falls short, so you can refine it. And because each lesson is one small rule, this loop could eventually run on its own, with tools that review agent sessions and propose rules.

  1. Run syllabus project init to create a .syllabus directory in your repo.
  2. In .syllabus/config.yaml, choose the skills and rules you want from libraries, and add any of your own.
  3. Run syllabus project sync. Syllabus copies in the rules, writes RULES.md to index them, and installs the skills where your agents look for them.
  4. Syllabus adds a Skills and Rules section to your AGENTS.md and keeps it current, so your agents read the rules and use the skills that fit each task.
  5. Commit the results. When a library publishes new versions, run syllabus project update to adopt them.

Set up your first project walks through these steps.

Syllabus is built around just a few simple concepts.

You write a single best practice as a rule, written as a Markdown file. A rule tells an agent what to do, when it applies, and how to apply it.

You organize rules into rule groups. There are two kinds of rule groups: technology groups, for a technology like Go, Tailwind, Next.js, Typescript, or Playwright, and practice groups, for a practice like testing, logging, or error handling.

A skill is the AI skill you already know, officially an Agent Skills folder: a reusable procedure, such as how to build a feature or review a pull request. Skills are coarse-grained, so a skill cites the rules it applies instead of restating them.

You declare the rules and skills you want in a project, specifically in the /.syllabus folder in your project repo. You (or your agent) can use the syllabus CLI to initialize a new project, and add your own rules, rule groups, and skills. We recommend committing /.syllabus, and the skills Syllabus installs, to version control.

Finally, you probably want to include some pre-existing rules and skills from a library, a repository that publishes rule groups and skills for other projects. A library can require or recommend skills for every project that uses it. You can also install skills from a Git repository that isn’t a library, such as a third party’s skill collection. Libraries can optionally declare license terms, which Syllabus carries into your project.

Syllabus combines your own rules and the ones you include from sources into guidance under .syllabus/generated/. Agents start with RULES.md, an index that tells them which rule groups to read.

An abbreviated RULES.md entry:

### Testing
**When to read this group:** When changing behavior or fixing bugs.
**Open group:** [Testing](groups/practices/testing.md)

The linked rule group page contains the rules in full, or links to their full definitions. For example, it might tell an agent to add a regression test when fixing a bug. See the generated file layout for details.

Syllabus writes each installed skill into .claude/skills/ and .agents/skills/, where coding agents find skills, and lists them in .syllabus/generated/SKILLS.md. Each installed skill that cites rules ends with links to your project’s version of every rule it applies. See Skill.

Syllabus assembles rules and skills from multiple sources into a cohesive, agent-friendly collection of files in your repo, and checks that those files are valid and current.

Syllabus installs skills into your repository, not on your computer. It doesn’t manage the personal skills a person installs for all of their repositories, though it warns when one has the same name as a skill your project installs.

Syllabus also supplies the instructions that tell your agents to follow the relevant rules and skills for each task: a Skills and Rules section, which tells them to read RULES.md and apply the rules that fit the task, and to use any installed skill listed in SKILLS.md that fits it. syllabus project init adds the section to AGENTS.md, and later commands keep it current between its markers without changing the rest of the file. To manage AGENTS.md yourself, turn the section off with manageAgentsFile: false. See Set up your first project.

That is where Syllabus’s work ends. It can’t guarantee that your agents followed your rules or used your skills. You can add your own guidance on how agents should apply them, in other procedures, in skills, or elsewhere in AGENTS.md.

Fabrica SyllabusMIT licensedPrivacy

Built with Fabrica.